The present invention relates to an architecture of a microcomputer, particularly a RISC (Reduced Instruction Set Computer) type microcomputer and, more particularly, to a technology effective if applied to a microcomputer to be packaged in a device for controlling it.
Moreover, the present invention relates to a circuit for coded division such as a dividing circuit for a coded binary number of arbitrary length and multi-precision and, more particularly, to a technology effective if applied to a step division of the RISC type microcomputer.
The most serious bottleneck for reducing the number of machine cycles necessary for executing one instruction is known to be the decoding of the instruction. In order to speed up this decoding, it is know effective to adopt an instruction format of fixed length so that where the boundary of the instruction resides may be informed before a preceding instruction has been interpreted. In the so-called "RISC type computer", most instructions are executed for one cycle by adopting the instruction format of fixed length and a pipe line of multiple steps. The conventional RISC computer has used a 32-bit instruction format without exception. This 32-bit fixed length instruction format is advantageous in that what register is to be read can be determined without decoding the operation code by fixing fields in the instruction formats of a source register and a destination register, and in that no alignment is required when an immediate value is decoded. On the contrary, the 32-bit fixed length instruction format requires 32 bits even no matter simple content of an instruction might be described. As a result, the number of bytes occupied by the instruction code is increased to raise a problem that the ratio of the memory area to be occupied by a program is accordingly increased. If the memory area occupied by the program is increased, a memory having a larger capacity has to be packaged to raise the cost of the microcomputer system, thus making it difficult to construct a system having an excellent performance ratio to the cost. Since the RISC processor is given an architecture for speeding up the execution of instructions by reducing the number of instructions, there arises a tendency that the undefined operation codes grow more for the instruction set. The multiplicity of the undefined operation codes deteriorates the code efficiency of the object program and degrades the memory using efficiency the worse.
The preceding patent application for improving such memory using efficiency or code efficiency is exemplified by Japanese Patent Application No. 222203/1990 (corresponding to U.S. patent application having Ser. No. 07/748,779 filed on Aug. 22, 1991). This application discloses a concept that the instruction format has a shorter bit number than that of the data word length. In this case, however, we have found that the various problems have to be solved by adopting the fixed length instruction format having a shorter bit number than the data word length. For example, new measures for the data processing have to be examined on the case, in which immediate data having a bit number equivalent to the data word length are necessary, or on the manner for assigning a branch destination address such as an absolute address as to the enlarged program or system structure. On the other hand, the above-specified application has failed to take any consideration into a relation of a power of 2 between the data word length and the instruction word length. Thus, the application has failed to positively prevent such a misalignment, in which one instruction is present across one word boundary of the memory, to leave new items such as the memory using efficiency, the software program simplification or the processing rate unexamined.
As a dividing technology to be executed in the microcomputer or the like, on the other hand, there is well known a division method, in which the codes of the quotient and the remainder are determined from the code of a dividend and the code of a divisor to execute the division with the absolute value of the dividend by a recovering method or a recovered method until the codes of the quotient and the remainder are finally corrected. In recent years, there are disclosed in the coded division several circuits and methods for executing the divisions in the coded state without taking the absolute values of the dividend and the divisor. In case the division is to be executed with the coded dividend and divisor, either method basically adopts the following procedures. Specifically, in case the code of the dividend or partial remainder and the code of the divisor are equal, the result of subtraction of the divisor from the dividend or partial remainder is used as a new partial remainder. In case, on the other hand, the code of the dividend or partial remainder and the code of the divisor are different, the result of an addition of the divisor to the dividend or partial remainder is used as a new partial remainder. Thus, the quotient is determined by repeating the subtractions or additions sequentially. At this time, in case the dividend is positive or in case the dividend is not contained by the divisor, a correct answer can be achieved by executing some quotient or remainder corrections on the basis of those procedures. In case, however, the dividend is negative and in case the dividend is contained by the divisor, the quotient thus determined is smaller than the correct quotient by the value "1" having an LSB weight toward the smaller absolute value. This error is caused by deeming the code of the partial remainder as correct in case the negative dividend or the partial remainder is subjected to the aforementioned addition or subtraction so that the partial remainder takes a zero.
In order to eliminate this error, there have been devised several dividing circuits which are equipped with means for detecting that the partial remainder is zero to correct the quotient. In Japanese Patent Laid-Open No. 165326/1990, for example, there is disclosed a technique, in which the irrecoverable dividing means is equipped with a register, which is set when the arithmetic result (i.e., the partial remainder) on each line is zero and reset when the value 1 enters the least significant bit on each line of the dividend, so that the quotient and remainder are corrected by using the result of the register. According to this disclosure, a correct coded division is realized by detecting and correcting the case, in which the partial remainder is zero, by using the aforementioned set and reset register. In Japanese Patent Laid-Open No. 171828/1990, on the other hand, there is disclosed another technique for preventing an erroneous quotient bit from being outputted in case the dividend is negative, by detecting whether or not the partial remainder is zero at each step of determining the quotient bit. In Japanese Patent Laid-Open No. 160235/1984, moreover, there is disclosed a technique which is equipped with a hardware for detecting the case, in which the partial remainder is zero, so that the most significant bit of the partial remainder may be deemed as 1 if the dividend is negative and if the partial remainder becomes zero in the course of the division.
Thus, in the prior art for the division with the coded dividend and divisor, the quotient bit is corrected by detecting that the partial remainder is zero. According to this technique, whether or not the partial remainder is zero has to be decided each time it is determined, and these decisions have to be accomplished n-times if the divisor has n bits. Moreover, whether or not the partial remainder is zero is not determined until all bits are examined. Therefore, the necessity for a special purpose hardware is anticipated if one decision is to be speeded up.